Analysis of Public Sentiment Towards the Discourse of the Gentengnization Program using the K-Nearst Neighbor Method

The discourse on the implementation of "tiling" policies—the standardization of residential roof materials and aesthetics—has sparked a wide debate in the digital public sphere, balancing the improvement of urban aesthetics with the financial burden of households. The study aimed to map the polarity distribution of public opinion into positive, negative, and neutral categories, as well as to evaluate the performance of the K-Nearest Neighbors (KNN) algorithm using Term Frequency-Inverse Document Frequency (TF-IDF) weighting under the CRISP-DM framework. Textual data collected from YouTube comments was processed through a pre-text processing stage, including purification, case folding, normalization, tokenization, stopword removal, and stemming. Lexicon-based labeling on 1,600 valid comments revealed 734 negative sentiments (45.88%), 699 positive sentiments (43.69%), and 167 neutral sentiments (10.44%), indicating that critical public sentiment remained dominant. This dataset was divided through an 80:20 stratification split into 1,180 training samples and 296 test samples. Model evaluation using K=15 with Euclidean distances yielded an overall accuracy of 70.95%. Detailed metrics showed strong detection for negative classes (69.70% precision, 78.23% recall, 73.72% F1 scores) and positive classes (73.08% accuracy, 66.43% recall, 69.60% F1 scores), but limitations in neutral classes (0.00% precision, 0.00% recall) caused by class imbalances. These findings provide empirical insights for policymakers to consider subsidies or material assistance before implementing urban roofing policies.